Automated sentinel garden monitoring system for early pest detection using computer vision and machine learning

Overview

This solution addresses a critical challenge for New Zealand's horticulture industry: the early detection of unwanted pests such as the Brown Marmorated Stink Bug (BMSB), which can infect multiple crops and cause widespread economic damage. Traditional pest monitoring relies on manually checked traps, which are costly, labour-intensive, and difficult to scale. This proposal offers an alternative through an automated sentinel garden monitoring system that combines computer vision, imaging analysis, and machine learning to detect pest presence and track spatial spread in near real time.

By deploying programmable cameras at key locations and processing images on the edge, the system reduces the need for expert field visits while improving the speed and accuracy of pest surveillance. The approach supports growers, industry bodies, and biosecurity agencies seeking practical, technology-driven tools to protect crop health and market access.

Technical specifications

• Edge-based image capture: Programmable cameras continuously collect images of sentinel plants, with sampling frequency tuned to balance detection sensitivity and data efficiency.
• On-device machine learning: A machine learning algorithm processes images locally at the edge, enabling rapid detection without requiring constant high-bandwidth data transmission.
• Species identification: When an intrusion is detected, captured images are analysed to identify the insect species, with the system designed to operate effectively using a limited number of training images.
• Remote connectivity and alerting: Detection events trigger automated signals sent to operators, enabling timely response.
• Networked garden design: A planned network of sentinel gardens supports optimal spatial coverage and monitoring of pest spread across a region.

Technology readiness level

The system is at an early-to-mid stage of development. Initial validation involves setting up a garden equipped with a fixed programmable camera and testing the machine learning pipeline for detection accuracy and species identification. Future plans include expanding from a single-camera setup to a networked configuration of sentinel gardens to optimise monitoring coverage. Further refinement of the machine learning algorithm and field validation across diverse conditions will be required before broader commercial deployment.